(1) Field of the Invention
The present invention relates to a multiprocessor system containing a plurality of data processing apparatuses and a networking apparatus which establishes physical links between the plurality of data processing apparatuses.
(2) Description of the Related Art
In some types of parallel processing systems containing a plurality of data processing apparatuses, a networking apparatus is provided for transferring data between the plurality of data processing apparatuses. The networking apparatus receives a connection request from one of the plurality of data processing apparatuses, and establishes a physical link between the data processing apparatus which sends the connection request to the networking apparatus, and another of the plurality of data processing apparatuses, which is requested by the connection request. Such a networking apparatus comprises a plurality of switch units. When trouble occurs in the networking apparatus, it is required that each data processing apparatus recognize the location of the trouble (in which switch unit the trouble has occurred), or recognize the data route which passes through the location of the trouble.
In conventional multiprocessor systems, a counter for detecting hang-up is provided in a data processing apparatus on a sender side for detecting hang-up of a data transfer between data processing apparatuses due to a fault in the networking apparatus. The counter counts the time elapsed from the start of the data transfer, and it determines that the networking apparatus is hung-up when the data transfer is not completed when the counter reaches a predetermined time.
However, the above method for detecting hang-up has the following drawbacks:
The data processing apparatus cannot recognize the location of the trouble (the cause of the hang-up). For example, a defective switch unit cannot be recognized by the data processing apparatus.
In addition, it takes a long time to detect the hang-up. For example, different values must be set for the above predetermined time, depending on the number of data processing apparatuses, the structure of the networking apparatus, and the like.
Further, since the location of the trouble cannot be recognized, a route for the next data transfer cannot be determined.